How xenbaloch/remdinet works
RemDiNet is a PyTorch low-light image enhancement framework. Its main workflow pairs low-light images with optional ground truth, trains a saliency/SNR-guided enhancement model through three loss stages, then applies the model to test images and evaluates enhancement quality. The map also shows color correction and metric calculation as internal capabilities. The README documents testing and stage behavior; unsampled source limits confirmation of some wiring, so only directly supported relationships are drawn.
Main entry points: lowlight_train.py Staged trainer · lowlight_test.py Test and inference · model.py SNR-saliency enhancer · dataloader.py Pairing and augmentation
Components 13
Data preparation
- Pairing and augmentation
dataloader.py
Enhancement model
- SNR-saliency enhancer
model.py - Color correction
color_enhancement.py
Training
- Staged trainer
lowlight_train.py - Enhancement losses
Myloss.py - Stage weights
loss_weights.py - Model checkpoint
Inference and evaluation
- Test and inference
lowlight_test.py - Enhanced images
- Image metrics
metrics.py - PSNR and MS-SSIM
metrics_core.py
Other components
- Researcher
- Image datasets
Connections 15
- Researcher → Staged trainer: starts training
- Image datasets → Pairing and augmentation: provides images
- Staged trainer → Pairing and augmentation: loads batches
- Pairing and augmentation → Staged trainer: returns batches
- Staged trainer → SNR-saliency enhancer: trains model
- Staged trainer → Enhancement losses: computes loss
- Enhancement losses → Stage weights: reads defaults
- Staged trainer → Model checkpoint: saves weights
- Researcher → Test and inference: starts inference
- Test and inference → SNR-saliency enhancer: enhances images
- Test and inference → Model checkpoint: loads weights
- SNR-saliency enhancer → Color correction: uses correction
- SNR-saliency enhancer → Enhanced images: produces images
- Test and inference → Image metrics: evaluates output
- Image metrics → PSNR and MS-SSIM: delegates metrics